|
|
|
Executive summary
Through June 30, 1998, the Supercomputing Institute will provide allocations to Minnesota
researchers of parallel vector processing (PVP) resources on the Cray C916/12512
and massively parallel processing (MPP) resources on the 272-node Cray T3E-900.
These allocations are available to faculty members at the University of Minnesota
and other researchers at accredited institutions of post-secondary education in the
state of Minnesota for their work and that of their students and research groups.
These resources are managed by the Minnesota Supercomputer Center Inc. (MSC Inc.),
which also provide user support. Peer review of allocations is the responsibility
of the Supercomputing Institute. As of July 1, 1998, the major supercomputing resources
will be manage by the Supercomputing Institute rather than MSC Inc. More information
about these new resources is available at
www.msi.umn.edu/general/Programs/Descriptions/ibm_sp.html
The contract with Minnesota Supercomputer Center, Inc. expires on June 30, 1998,
and Cray C90 and T3E-900 resources will not be available after that date.
Eligibility for allocations
All proposals must be sponsored and signed by a faculty member of the University
of Minnesota or other accredited institution of post-secondary education in the state
of Minnesota. Allocations are awarded on the basis of competitive review.
PVP resources
The Supercomputing Institute will provide through June 30, 1998 allocations of parallel
vector processing (PVP) resources on the Cray Y-MP C916/12512, which is called the
C90 in the rest of this call.
The C90 has a total of twelve central processing units and contains 512 megawords
of shared memory. The C90 system has a clock speed of under 4.2 nanoseconds and is
equipped with a 512 megaword Solid State storage Device (SSD). (The SSD is used by
the operating system to improve performance of input/output operations and throughput.)
The C90 processors each contain two sets of vector functional units: one for processing
the even-numbered vector elements and one for processing the odd-numbered vector
elements.
PVP resources are charged according to the following service unit (SU) formula:
MAX[0, MW - 64]
SUs = F * H ( 1 + ---------------- ) + C * S + D * SYS
A * 145
where:
SUs = service units
F = speed factor of 3.36 for C90
H = central-processor-unit hours
(through batch queue or interactive)
A = average number of central processor units used concurrently
C = 1.3889 x 10-5
S = a (usually small) add-on for megabytes transferred
D = 1.3889 x 10-8
MW = main memory used in megawords
SYS = operating system calls issued
In addition to the SU charges for central-processor-unit time, accounts will be
charged SUs for certain peripheral resources, including disk, printing, etc. A complete
list of charges with the current charging scheme is available in the Rate Sheet in
the Supercomputing Institute Computer Users Package at http://www/msci/rate_sheet.html.
To facilitate the work of the Institute’s researchers and to ensure full utilization
of the resources purchased from MSC Inc., the Institute has developed a reduced-priority
queue for PVP resources. This will be called the p queue, and details will be announced
soon. To apply for these resources please check the appropriate box on the request
form.
MPP resources
The Supercomputing Institute will provide allocations of massively parallel processing
(MPP) resources on the 256-node Cray T3E-900 massively parallel computer. MPP resources
will be charged according to the following service unit (SU) formula:
SUs = (1 + 0.30 * MAX[0, (MB - 117)/117]) * H * PE * T
where:
SUs = service units
H = Number of hours at least one MPP system processing element was
used or reserved
MB = Millions of bytes used per PE
PE = Number of MPP system processing elements used or reserved
T = 0.7826 (if T3E)
Each of the 256 processors on the T3E has a peak processing rate of 900 megaflops.
The T3E-900 processors run at 450 MHz (compared to 300 MHz in previous T3E system
and 150 MHz in the previous T3D system). In addition to fast processors, these systems
feature a latency-hiding memory interface and a high-performance interprocessor interconnect.
The interprocessor network's 3-D bi-directional torus design helps to minimize the
number of hops between nodes and provides excellent scalability up to a maximum configuration
of 2048 processing elements (PEs). MPP SUs may also be used for (and are required
for) peripheral charges—as discussed above.
Summary: allocation of resources
Allocations are made to a specific resource classification (i.e., PVP or MPP) and
are for a fixed number of service units (SUs). SUs may be used for disk and printing,
as required, as well as for actual supercomputer resources. One service unit is approximately
equivalent to one-third of an hour of supercomputing time for low-memory jobs on
the C90 or 1.28 processor element hours on the T3E-900.
Completing projects utilizing Cray C90 and Cray T3E-900
resources
We encourage researchers who have projects they want to complete prior to the
transition from Cray resources on June 30, 1998 to utilize the resources that have
been allocated to them and, if they run out of resources, to apply for an additional
allocation of resources. Currently, C90 resources are available for allocation.
We are not at this time making additional allocation of Cray T3E-900 resources;
depending, however, on usage patterns this may change. To apply for an additional
allocation of resources on either the C90 or the T3E-900, the Principal Investigator
should submit a request which i) describes the progress that has been made to date,
ii) discusses the work that is proposed, iii) states and justifies the amount of
resources needed, and iv) discusses any special circumstance (e.g., publication deadlines,
upcoming travel). Ordinarily each of the first two items should be at most one paragraph,
and the fourth item is optional. These requests, which will be peer reviewed, should
be submitted to Ms. Kay Anderson, the Institute’s Research Programs Administrator,
in care of the Institute.
Request justification and review criteria
Supercomputing Institute allocation requests should be justified in terms
of their usage for promoting the mission of the Supercomputing Institute, which is
supercomputing research.
In reviewing proposals and awarding allocations the Supercomputing Institute Peer
Review and Allocation Panel is asked to consider the quality of the proposed research,
the need for supercomputing resources and the efficiency of use of those resources,
the principal investigator’s ability to carry out the work as judged by past accomplishments,
external funding, or other evidence of successful peer review or future potential,
the principal investigator’s past usage of supercomputing resource allocations and
his or her contributions, if any, to the UMSI Research Report Series, the importance
of the proposed work in its own field, opportunities for leverage, and the relevance
of the work to the University’s missions of education, research, and technology transfer.
|
![[Program Information]](http://www.msi.umn.edu/general/buttons/Alloc&Grants.GIF)
![[Introduction to the Institute]](http://www.msi.umn.edu/general/buttons/InstIntro.GIF)
![[Supercomputing Institute Homepage]](http://www.msi.umn.edu/general/buttons/MSIhome.GIF)